Table 2.

List of aforementioned AI algorithms showing applications to bacterial classification via Raman spectroscopy along with corresponding methodology and considerations for each algorithm.

AI Algorithm	Target Organism	Accuracy	Methodology	Considerations	Reference
Support Vector Machince (SVM)	Escherichia coli	81.1%	Use hyperplane optimization to demarcate between class data	Not inherently designed for multi-class (2+) classification	[126,128]
Random Forests (RFs)	3 bacterial and 3 archaeal species	98.9%	Average of multiple decision trees trained on random subsets of training data	Lack of interpretability and tendency to overfit model	[133,134]
k-nearest-neighbors (KNN)	10 methicillin-resistant S. aureus, 6 methicillin-sensitive S. aureus, and 6 L. pneumophila isolates	97.8%	Maps high dimensional data to a higher dimensional space and define class members based on proximity by a distance measure	Optimization of k along with computational complexity requires extended effort	[139,140]
Gradient Boosted Machines (GBM)	15 strains of Klebsiella pneumoniae based on Carbapenem resistance	99.40%	Apply loss function to a base learner (decision tree, regression model, etc.) and repeat training until loss function reaches minima	Computational complexity due to number of iterations needed to minimize loss function	[137]
Convolutional Neural Networks (CNN)	30 species and strains of various bacteria	89.1%	Model neuronal connections based on activation function for input classification	Complex theory behind neural networks requires expert knowledge before use	[119]